Approach to Daily Load Forecast of VSNN Based on Data Mining

Abstract

The keys of improving the precision of daily load forecasting lie in the fore processing and the forecasting model, so this paper puts forward a new method of vary structure neural network (shorten as "VSNN") for power load forecast which is based on united data mining technology. Firstly, to search the historical daily load which have the same meteorological category as the forecasting day; secondly, to make further collection of data to compose data sequence with highly similar meteorological features which can boost up rules and weaken disturbance; thirdly, to constitute VSNN forecasting model accordingly. So the model can overcome the disadvantages of ANN through vary structure optimization to determine the optimal structure and optimal fitting approximation, and it does not easily convergence, not easily trap in partial minimum, and its structure can be determined by itself not by artificially. In the end, the forecasting precision was improved effectively, the input and calculation model was simplified properly, and the software programming was easier to realize. So the new method is more practical.